Re: [PATCH v5 tip/core/locking 6/7] locking: Add ansmp_mb__after_unlock_lock() for UNLOCK+LOCK barrier

[Oleg Nesterov]

On 12/10, Paul E. McKenney wrote:
>
> On Tue, Dec 10, 2013 at 06:04:04PM +0100, Oleg Nesterov wrote:
> >
> > I am wondering, perhaps smp_mb__after_unlock() makes more sense?
> >
> > Note that it already has the potential user:
> >
> > 	--- x/kernel/sched/wait.c
> > 	+++ x/kernel/sched/wait.c
> > 	@@ -176,8 +176,9 @@ prepare_to_wait(wait_queue_head_t *q, wa
> > 		spin_lock_irqsave(&q->lock, flags);
> > 		if (list_empty(&wait->task_list))
> > 			__add_wait_queue(q, wait);
> > 	-	set_current_state(state);
> > 	+	__set_current_state(state);
> > 		spin_unlock_irqrestore(&q->lock, flags);
> > 	+	smp_mb__after_unlock();
> > 	 }
> > 	 EXPORT_SYMBOL(prepare_to_wait);
> >
> > 	@@ -190,8 +191,9 @@ prepare_to_wait_exclusive(wait_queue_hea
> > 		spin_lock_irqsave(&q->lock, flags);
> > 		if (list_empty(&wait->task_list))
> > 			__add_wait_queue_tail(q, wait);
> > 	-	set_current_state(state);
> > 	+	__set_current_state(state);
> > 		spin_unlock_irqrestore(&q->lock, flags);
> > 	+	smp_mb__after_unlock();
> > 	 }
> > 	 EXPORT_SYMBOL(prepare_to_wait_exclusive);
> >
> >
> > Assuming it can also be used "later", after another LOCK, like in
> > your example in 5/7.
>
> I am fine either way.  But there was an objection to tying this to the
> unlock because it costs more on many architectures than tying this to
> the lock.

OK, I see, thanks.

> But if you are saying "in addition to" rather than "instead of" that
> would be a different conversation.

Yes, please forget for now.

Oleg.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/